Multi-Area, Multi-Service and Multi-Tier Edge-Cloud Continuum Planning

This paper presents the optimal planning of multi-area, multi-service, and multi-tier edge-cloud environments. The goal is to evaluate the regional deployment of the compute continuum, i.e., the type and number of processing devices, their pairing with a specific tier and task among different areas subject to processing, rate, and latency requirements. Different offline compute continuum planning approaches are investigated and detailed analysis related to various design choices is depicted. We study one scheme using all tasks at once and two others using smaller task batches. The latter both iterative schemes finish once all task groups have been traversed. Group-based approaches are presented as dealing with potentially excessive execution times for real-world sized problems. Solutions are provided for continuum planning using both direct complex and simpler, faster methods. Results show that processing all tasks simultaneously yields better performance but requires longer execution, while medium-sized batches achieve good performance faster. Thus, the batch-oriented schemes are capable of handling larger problem sizes. Moreover, the task selection strategy in group-based schemes influences the performance. A more detailed analysis is performed in the latter case, and different clustering methods are also considered. Based on our simulations, random selection of tasks in group-based approaches achieves better performance in most cases.